MutliPV behavior?

[Ovyron]

In SMP mode, I want to do a full PV search for all root moves,
eventually splitting the root moves over all helper threads.

What's the difference between this and setting MultiPV to the maximum?

[syzygy]

In SMP mode, I want to do a full PV search for all root moves,
eventually splitting the root moves over all helper threads.

What's the difference between this and setting MultiPV to the maximum?

Setting MultiPV to the maximum performs a regular full search on all moves to find the best move, then a regular full search on all moves minus the best one to find the second best move, etc.

This is easy to implement, but seems suboptimal. Of course the transposition table helps to avoid double work.

[Joerg Oster]

In SMP mode, I want to do a full PV search for all root moves,
eventually splitting the root moves over all helper threads.

What's the difference between this and setting MultiPV to the maximum?

The difference is I don't switch the main search thread into MultiPV mode, but only the helper threads.

First tests indicate that I eventually need to modify the TT probing, too.
The helper threads already showed a mate score on a test position,
yet the main thread didn't change to that move ...

[Ovyron]

Thanks Joerg!

You're welcome!
Thanks for bringing this up.

You're welcome too!

Did your patch make it to Stockfish 9?

[Joerg Oster]

Thanks Joerg!

You're welcome!
Thanks for bringing this up.

You're welcome too!

Did your patch make it to Stockfish 9?

No, as I didn't create a pull request for it.
But I just started a discussion thread on this in the forum to get some feedback.